Merge branch 'release/v0.3.1' into develop

api/app/services/api_key_service.py

@@ -51,6 +51,19 @@ class ApiKeyService:
             if existing:
                 raise BusinessException(f"API Key 名称 {data.name} 已存在", BizCode.API_KEY_DUPLICATE_NAME)
 
+            # 若 rate_limit 超过租户套餐的 api_ops_rate_limit，直接报错
+            from app.models.workspace_model import Workspace
+            from app.core.quota_manager import get_api_ops_rate_limit
+
+            workspace = db.query(Workspace).filter(Workspace.id == workspace_id).first()
+            if workspace:
+                tenant_api_ops_limit = get_api_ops_rate_limit(db, workspace.tenant_id)
+                if tenant_api_ops_limit and data.rate_limit > tenant_api_ops_limit:
+                    raise BusinessException(
+                        f"API Key QPS 不能超过套餐上限 {tenant_api_ops_limit}",
+                        BizCode.BAD_REQUEST
+                    )
+
             # 生成 API Key
             api_key = generate_api_key(data.type)
 
@@ -152,6 +165,20 @@ class ApiKeyService:
             if existing:
                 raise BusinessException(f"API Key 名称 {data.name} 已存在", BizCode.API_KEY_DUPLICATE_NAME)
 
+        # 若 rate_limit 超过租户套餐的 api_ops_rate_limit，直接报错
+        if data.rate_limit is not None:
+            from app.models.workspace_model import Workspace
+            from app.core.quota_manager import get_api_ops_rate_limit
+
+            workspace = db.query(Workspace).filter(Workspace.id == workspace_id).first()
+            if workspace:
+                tenant_api_ops_limit = get_api_ops_rate_limit(db, workspace.tenant_id)
+                if tenant_api_ops_limit and data.rate_limit > tenant_api_ops_limit:
+                    raise BusinessException(
+                        f"API Key QPS 不能超过套餐上限 {tenant_api_ops_limit}",
+                        BizCode.BAD_REQUEST
+                    )
+
         update_data = data.model_dump(exclude_unset=True)
         ApiKeyRepository.update(db, api_key_id, update_data)
         db.commit()
@@ -248,42 +275,14 @@ class RateLimiterService:
     def __init__(self):
         self.redis = aio_redis
 
-    async def check_tenant_rate_limit(self, tenant_id: uuid.UUID, limit: int) -> Tuple[bool, dict]:
-        """
-        按 tenant_id 做 1 秒滑动窗口限速，限制值来自套餐配额 api_ops_rate_limit
-        """
-        now = time.time()
-        window_start = now - 1  # 1 秒窗口
-        key = f"rate_limit:tenant_qps:{tenant_id}"
-
-        async with self.redis.pipeline() as pipe:
-            # 清理 1 秒前的旧记录
-            pipe.zremrangebyscore(key, 0, window_start)
-            # 加入当前请求（score=时间戳，member=时间戳+随机数保证唯一）
-            pipe.zadd(key, {f"{now}:{uuid.uuid4().hex}": now})
-            # 统计窗口内请求数
-            pipe.zcard(key)
-            # 设置 key 过期（2 秒后自动清理）
-            pipe.expire(key, 2)
-            results = await pipe.execute()
-
-        current = results[2]
-        remaining = max(0, limit - current)
-        reset_time = int(now) + 1
-
-        return current <= limit, {
-            "limit": limit,
-            "remaining": remaining,
-            "reset": reset_time,
-        }
-
     async def check_qps(self, api_key_id: uuid.UUID, limit: int) -> Tuple[bool, dict]:
-        """
-        检查QPS限制
+        """检查QPS限制
+
         Returns:
             (is_allowed, rate_limit_info)
         """
         key = f"rate_limit:qps:{api_key_id}"
+
         async with self.redis.pipeline() as pipe:
             pipe.incr(key)
             pipe.expire(key, 1, nx=True)  # 1 秒过期
@@ -295,8 +294,9 @@ class RateLimiterService:
 
         return current <= limit, {
             "limit": limit,
+            "current": current,
             "remaining": remaining,
-            "reset": reset_time
+            "reset": reset_time,
         }
 
     async def check_daily_requests(
@@ -304,7 +304,9 @@ class RateLimiterService:
             api_key_id: uuid.UUID,
             limit: int
     ) -> Tuple[bool, dict]:
-        """检查日调用量限制"""
+        """检查日调用量限制。
+        使用原子 INCR，先写后判断，极低概率下允许轻微超限（并发场景下可接受）。
+        """
         today = datetime.now().strftime("%Y%m%d")
         key = f"rate_limit:daily:{api_key_id}:{today}"
 
@@ -313,6 +315,7 @@ class RateLimiterService:
             hour=0, minute=0, second=0, microsecond=0
         )
         expire_seconds = int((tomorrow_0 - now).total_seconds())
+        reset_time = int(tomorrow_0.timestamp())
 
         async with self.redis.pipeline() as pipe:
             pipe.incr(key)
@@ -320,36 +323,74 @@ class RateLimiterService:
             results = await pipe.execute()
 
         current = results[0]
-        remaining = max(0, limit - current)
-        reset_time = int(tomorrow_0.timestamp())
 
-        return current <= limit, {
+        if current > limit:
+            return False, {
+                "limit": limit,
+                "remaining": 0,
+                "reset": reset_time,
+            }
+
+        return True, {
             "limit": limit,
-            "remaining": remaining,
-            "reset": reset_time
+            "remaining": max(0, limit - current),
+            "reset": reset_time,
         }
 
     async def check_all_limits(
             self,
-            api_key: ApiKey
+            api_key: ApiKey,
+            db: Optional[Session] = None,
     ) -> Tuple[bool, str, dict]:
         """
-        检查所有限制
-        Returns:
-            (is_allowed, error_message, rate_limit_headers)
+        检查所有限制，按以下顺序：
+        1. API Key QPS：取 api_key.rate_limit 与套餐 api_ops_rate_limit 的最小值作为限额
+        2. API Key 日调用量
         """
-        # Check QPS
-        qps_ok, qps_info = await self.check_qps(
-            api_key.id,
-            api_key.rate_limit
-        )
+        # 1. 取套餐限额与 api_key 自身限额的最小值
+        effective_limit = api_key.rate_limit
+        if db is not None:
+            try:
+                from app.models.workspace_model import Workspace
+                from app.core.quota_manager import get_api_ops_rate_limit
+
+                cache_key = f"tenant_api_ops_limit:{api_key.workspace_id}"
+                cached = await self.redis.get(cache_key)
+                if cached is not None:
+                    try:
+                        tenant_limit = int(cached) if cached != "0" else None
+                    except (ValueError, TypeError):
+                        cached = None
+                        tenant_limit = None
+
+                if cached is None:
+                    workspace = db.query(Workspace).filter(Workspace.id == api_key.workspace_id).first()
+                    if workspace:
+                        tenant_limit = get_api_ops_rate_limit(db, workspace.tenant_id)
+                        await self.redis.set(cache_key, str(tenant_limit) if tenant_limit else "0", ex=60)
+                    else:
+                        tenant_limit = None
+
+                if tenant_limit:
+                    effective_limit = min(api_key.rate_limit, tenant_limit)
+            except Exception as e:
+                logger.warning(f"获取套餐限额失败，使用 api_key 自身限额: {e}")
+
+        # 用最终有效限额做 QPS 检查
+        qps_ok, qps_info = await self.check_qps(api_key.id, effective_limit)
         if not qps_ok:
-            return False, "QPS limit exceeded", {
+            # 判断是套餐限额触发还是 api_key 自身限额触发
+            if tenant_limit and effective_limit == tenant_limit and api_key.rate_limit > tenant_limit:
+                error_msg = "Tenant limit exceeded"
+            else:
+                error_msg = "QPS limit exceeded"
+            return False, error_msg, {
                 "X-RateLimit-Limit-QPS": str(qps_info["limit"]),
                 "X-RateLimit-Remaining-QPS": str(qps_info["remaining"]),
                 "X-RateLimit-Reset": str(qps_info["reset"])
             }
 
+        # 2. 检查日调用量
         daily_ok, daily_info = await self.check_daily_requests(
             api_key.id,
             api_key.daily_request_limit
@@ -361,14 +402,13 @@ class RateLimiterService:
                 "X-RateLimit-Reset": str(daily_info["reset"])
             }
 
-        headers = {
+        return True, "", {
             "X-RateLimit-Limit-QPS": str(qps_info["limit"]),
             "X-RateLimit-Remaining-QPS": str(qps_info["remaining"]),
             "X-RateLimit-Limit-Day": str(daily_info["limit"]),
             "X-RateLimit-Remaining-Day": str(daily_info["remaining"]),
-            "X-RateLimit-Reset": str(daily_info["reset"])
+            "X-RateLimit-Reset": str(daily_info["reset"]),
         }
-        return True, "", headers
 
 
 class ApiKeyAuthService:

api/app/services/app_dsl_service.py

@@ -434,19 +434,37 @@ class AppDslService:
     def _resolve_model(self, ref: Optional[dict], tenant_id: uuid.UUID, warnings: list) -> Optional[uuid.UUID]:
         if not ref:
             return None
-        q = self.db.query(ModelConfig).filter(
-            ModelConfig.tenant_id == tenant_id,
-            ModelConfig.name == ref.get("name"),
-            ModelConfig.is_active.is_(True)
-        )
-        if ref.get("provider"):
-            q = q.filter(ModelConfig.provider == ref["provider"])
-        if ref.get("type"):
-            q = q.filter(ModelConfig.type == ref["type"])
-        m = q.first()
-        if not m:
-            warnings.append(f"模型 '{ref.get('name')}' 未匹配，已置空，请导入后手动配置")
-        return m.id if m else None
+        model_id = ref.get("id")
+        if model_id:
+            try:
+                model_uuid = uuid.UUID(str(model_id))
+                m = self.db.query(ModelConfig).filter(
+                    ModelConfig.id == model_uuid,
+                    ModelConfig.tenant_id == tenant_id,
+                    ModelConfig.is_active.is_(True)
+                ).first()
+                if m:
+                    return str(m.id)
+            except (ValueError, AttributeError):
+                pass
+        model_name = ref.get("name")
+        if model_name:
+            q = self.db.query(ModelConfig).filter(
+                ModelConfig.tenant_id == tenant_id,
+                ModelConfig.name == model_name,
+                ModelConfig.is_active.is_(True)
+            )
+            if ref.get("provider"):
+                q = q.filter(ModelConfig.provider == ref["provider"])
+            if ref.get("type"):
+                q = q.filter(ModelConfig.type == ref["type"])
+            m = q.first()
+            if m:
+                return str(m.id)
+            warnings.append(f"模型 '{model_name}' 未匹配，已置空，请导入后手动配置")
+        else:
+            warnings.append(f"模型 ID '{model_id}' 未匹配，已置空，请导入后手动配置")
+        return None
 
     def _resolve_kb(self, ref: Optional[dict], workspace_id: uuid.UUID, warnings: list) -> Optional[str]:
         if not ref:
@@ -587,7 +605,7 @@ class AppDslService:
                     if not kb_id:
                         continue
                     kb_ref = {}
-                    if isinstance(kb_id, str) and len(kb_id) >= 36:
+                    if isinstance(kb_id, str):
                         try:
                             uuid.UUID(kb_id)
                             kb_ref["id"] = kb_id
@@ -601,6 +619,33 @@ class AppDslService:
                     else:
                         warnings.append(f"[{node_label}] 知识库 '{kb_id}' 未匹配，已移除，请导入后手动配置")
                 config["knowledge_bases"] = resolved_kbs
+            elif node_type in (NodeType.LLM.value, NodeType.QUESTION_CLASSIFIER.value, NodeType.PARAMETER_EXTRACTOR.value):
+                model_ref = config.get("model_id")
+                if model_ref:
+                    ref_dict = None
+                    if isinstance(model_ref, dict):
+                        ref_id = model_ref.get("id")
+                        ref_name = model_ref.get("name")
+                        if ref_id:
+                            ref_dict = {"id": ref_id}
+                        elif ref_name is not None:
+                            ref_dict = {"name": ref_name, "provider": model_ref.get("provider"), "type": model_ref.get("type")}
+                    elif isinstance(model_ref, str):
+                        try:
+                            uuid.UUID(model_ref)
+                            ref_dict = {"id": model_ref}
+                        except ValueError:
+                            ref_dict = {"name": model_ref}
+                    if ref_dict:
+                        resolved_model_id = self._resolve_model(ref_dict, tenant_id, warnings)
+                        if resolved_model_id:
+                            config["model_id"] = resolved_model_id
+                        else:
+                            warnings.append(f"[{node_label}] 模型未匹配，已置空，请导入后手动配置")
+                            config["model_id"] = None
+                    else:
+                        warnings.append(f"[{node_label}] 模型未匹配，已置空，请导入后手动配置")
+                        config["model_id"] = None
             resolved_nodes.append({**node, "config": config})
         return resolved_nodes
 

api/app/services/knowledge_service.py

@@ -7,7 +7,6 @@ from app.models.models_model import ModelConfig
 from app.schemas.knowledge_schema import KnowledgeCreate, KnowledgeUpdate
 from app.repositories import knowledge_repository
 from app.core.logging_config import get_business_logger
-from app.repositories.model_repository import ModelConfigRepository
 from app.models.models_model import ModelType
 
 business_logger = get_business_logger()
@@ -78,41 +77,31 @@ def create_knowledge(
         tenant_id = workspace.tenant_id
 
         if not knowledge.embedding_id:
-            embedding_models = ModelConfigRepository.get_by_type(
-                db=db, model_types=[ModelType.EMBEDDING], tenant_id=tenant_id, is_active=True
-            )
-            if embedding_models:
-                knowledge.embedding_id = embedding_models[0].id
-                business_logger.debug(f"Auto-bind embedding model: {embedding_models[0].id}")
+            if not workspace.embedding:
+                raise Exception("工作空间未配置 Embedding 模型，请先完善工作空间配置后重试")
+            knowledge.embedding_id = workspace.embedding
 
         if not knowledge.reranker_id:
-            rerank_models = ModelConfigRepository.get_by_type(
-                db=db, model_types=[ModelType.RERANK], tenant_id=tenant_id, is_active=True
-            )
-            if rerank_models:
-                knowledge.reranker_id = rerank_models[0].id
-                business_logger.debug(f"Auto-bind rerank model: {rerank_models[0].id}")
+            if not workspace.rerank:
+                raise Exception("工作空间未配置 Rerank 模型，请先完善工作空间配置后重试")
+            knowledge.reranker_id = workspace.rerank
 
         if not knowledge.llm_id:
-            llm_models = ModelConfigRepository.get_by_type(
-                db=db, model_types=[ModelType.LLM, ModelType.CHAT], tenant_id=tenant_id, is_active=True
-            )
-            if llm_models:
-                knowledge.llm_id = llm_models[0].id
-                business_logger.debug(f"Auto-bind llm model: {llm_models[0].id}")
+            if not workspace.llm:
+                raise Exception("工作空间未配置 LLM 模型，请先完善工作空间配置后重试")
+            knowledge.llm_id = workspace.llm
 
         if not knowledge.image2text_id:
-            image2text_models = db.query(ModelConfig).filter(
+            model = db.query(ModelConfig).filter(
                 ModelConfig.tenant_id == tenant_id,
-                ModelConfig.type.in_([ModelType.CHAT.value]),
+                ModelConfig.type.in_([ModelType.CHAT.value, ModelType.LLM.value]),
                 ModelConfig.capability.contains(["vision"]),
                 ModelConfig.is_active == True,
-                ModelConfig.is_composite == False
-            ).order_by(ModelConfig.created_at.desc()).all()
-            if not image2text_models:
+            ).order_by(ModelConfig.created_at.desc()).first()
+            if not model:
                 raise Exception("租户下没有可用的视觉模型，创建知识库失败")
-            knowledge.image2text_id = image2text_models[0].id
-            business_logger.debug(f"Auto-bind image2text model: {image2text_models[0].id}")
+            knowledge.image2text_id = model.id
+            business_logger.debug(f"Auto-bind image2text model: {model.id}")
 
         business_logger.debug(f"Start creating the knowledge base: {knowledge.name}")
         db_knowledge = knowledge_repository.create_knowledge(

api/app/services/memory_agent_service.py

@@ -1282,7 +1282,7 @@ def get_end_user_connected_config(end_user_id: str, db: Session) -> Dict[str, An
     }
 
     logger.info(
-        f"Successfully retrieved connected config: memory_config_id={memory_config_id}, workspace_id={app.workspace_id}")
+        f"Successfully retrieved connected config: memory_config_id={memory_config_id}, workspace_id={end_user.workspace_id}")
     return result
 
 

api/app/services/model_service.py

@@ -125,11 +125,7 @@ class ModelConfigService:
                 api_key=api_key,
                 base_url=api_base,
                 is_omni=is_omni,
-                capability=capability,
-                extra_params={
-                    "temperature": 0.7,
-                    "max_tokens": 100
-                }
+                capability=capability
             )
 
             # 根据模型类型选择不同的验证方式
@@ -373,6 +369,15 @@ class ModelConfigService:
                 raise BusinessException("模型名称已存在", BizCode.DUPLICATE_NAME)
 
         model = ModelConfigRepository.update(db, model_id, model_data, tenant_id=tenant_id)
+
+        # 同步更新关联 api_keys 的 capability 和 is_omni
+        if model_data.capability is not None or model_data.is_omni is not None:
+            for api_key in model.api_keys:
+                if model_data.capability is not None:
+                    api_key.capability = model_data.capability
+                if model_data.is_omni is not None:
+                    api_key.is_omni = model_data.is_omni
+
         db.commit()
         db.refresh(model)
         return model